Two pricing concepts
- Live settlement: models and billable rates actually available through current channels. Treat these as the source of truth.
- Reference: rows shown to compare with official public prices. They are not callable offers and not the final bill.
How a charge is built
Charge = usage × group multiplier × the price snapshot used for that request. When reading live pricing, check:
- input / output token rates
- cache-write and cache-read rates (some models distinguish them)
- per-request models (for example some image models)
- the active group multiplier
Reading usage
The Usage Logs page records model, token details, cache usage, multiplier, route, duration, and final charge for every request. It is the single source of truth for verifying a bill.
Why one action can create multiple records
Agent clients may call a main model, helper models, compaction, or token-counting endpoints, and routing retries create separate records too. Compare request ids and timestamps before attributing the cause.
Top-ups
Top up on the Wallet page; charges are applied in real time per request (pre-charge then settle). Requests are rejected with 403 when balance is insufficient.
Referral credit
Referral rewards are USD-denominated API credit for eligible usage. They are not withdrawable cash.